Behavioral Sleep Medicine
ISSN: 1540-2002 (Print) 1540-2010 (Online) Journal homepage: http://www.tandfonline.com/loi/hbsm20
Relationships Between Questionnaire Ratings of Sleep Quality and Polysomnography in Healthy Adults Anna Westerlund, Ylva Trolle Lagerros, Göran Kecklund, John Axelsson & Torbjörn Åkerstedt To cite this article: Anna Westerlund, Ylva Trolle Lagerros, Göran Kecklund, John Axelsson & Torbjörn Åkerstedt (2014): Relationships Between Questionnaire Ratings of Sleep Quality and Polysomnography in Healthy Adults, Behavioral Sleep Medicine, DOI: 10.1080/15402002.2014.974181 To link to this article: http://dx.doi.org/10.1080/15402002.2014.974181
Published online: 10 Nov 2014.
Submit your article to this journal
Article views: 79
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=hbsm20 Download by: [New York University]
Date: 16 September 2015, At: 12:37
Downloaded by [New York University] at 12:37 16 September 2015
Behavioral Sleep Medicine, 00:1–15, 2014 Copyright © Taylor & Francis Group, LLC ISSN: 1540-2002 print/1540-2010 online DOI: 10.1080/15402002.2014.974181
Relationships Between Questionnaire Ratings of Sleep Quality and Polysomnography in Healthy Adults Anna Westerlund Clinical Epidemiology Unit, Department of Medicine Karolinska Institutet
Ylva Trolle Lagerros Clinical Epidemiology Unit, Department of Medicine Karolinska Institutet
Göran Kecklund Stress Research Institute, Stockholm University Behavioural Science Institute, Radboud University Division of Psychology, Department of Clinical Neuroscience Karolinska Institutet
John Axelsson Division of Psychology, Department of Clinical Neuroscience Karolinska Institutet
Torbjörn Åkerstedt Stress Research Institute, Stockholm University Division of Psychology, Department of Clinical Neuroscience Karolinska Institutet
This study aimed to examine the association between polysomnographic sleep and subjective habitual sleep quality and restoration from sleep. Thirty-one normal sleepers completed the Karolinska Sleep Questionnaire and multiple home polysomnography recordings (n D 2–5). Using linear regression, sleep quality and restoration were separately analyzed as functions of standard Correspondence should be addressed to Anna Westerlund, Karolinska Institutet, Department of Medicine/Solna, Clinical Epidemiology Unit, Eugeniahemmet T2, 171 76, Stockholm, Sweden. E-mail: [email protected]
1
2
WESTERLUND ET AL.
Downloaded by [New York University] at 12:37 16 September 2015
polysomnography parameters: sleep efficiency, total sleep time, sleep latency, stage 1 and 2 sleep, slow-wave sleep, rapid eye movement sleep, wake time after sleep onset, and awakenings (n), averaged across recordings. Stage 2 and slow-wave sleep predicted worse and better sleep quality, respectively. Also, slow-wave sleep predicted less subjective restoration, although adjustment for age attenuated this relation. Our findings lend some physiological validity to ratings of habitual sleep quality in normal sleepers. Data were less supportive of a physiological correlate of ratings of restoration from sleep.
With an ever-growing literature on the link between sleep habits and health outcomes (Ayas et al., 2003; Chandola, Ferrie, Perski, Akbaraly, & Marmot, 2010; Clark, Lange, Hallqvist, Jennum, & Rod, 2014; Gangwisch et al., 2006; Gangwisch, Malaspina, Boden-Albala, & Heymsfield, 2005; Ikehara et al., 2009; Kripke, Garfinkel, Wingard, Klauber, & Marler, 2002; Yaggi, Araujo, & McKinlay, 2006), simple tools to assess habitual sleep are being frequently utilized. Yet, studies on the association of subjective reports of sleep quality and sleep complaints with polysomnographic sleep are scarce. Consequently, validity assessments of epidemiological investigations are also impaired. Habitual sleep is typically measured with single time-point self-administered questionnaires assessing the experience of sleep over some specified period of time. The assumption is that such a report of sleep quality or complaints reflects a trait-like or a long-term state. Examples of self-administered sleep questionnaires include the Pittsburgh Sleep Quality Index (PSQI; Buysse, Reynolds, Monk, Berman, & Kupfer, 1989), the Basic Nordic Sleep Questionnaire (Partinen & Gislason, 1995), the Sleep Quality Scale (Yi, Shin, & Shin, 2006), and the Karolinska Sleep Questionnaire (Kecklund & Åkerstedt, 1992), which were developed to retrospectively assess sleep quality and complaints in various adult populations. Because these questionnaires reflect the subjective experience of sleep, they would likely benefit from validation against polysomnographic (objective) sleep. This has, though, rarely been done. One exception is the validation study of the PSQI among subjects without sleep complaints and depressed or sleep-disorder patients (Buysse et al., 1989). A weak positive correlation between global PSQI scores (a higher score reflecting worse sleep quality) and sleep latency according to polysomnography (PSG) was demonstrated for all subjects. In those with no sleep complaints, the global score positively correlated with percentage sleep time spent in rapid eye movement (REM) sleep. With regard to sleep scales for the assessment of habitual sleep in nonclinical populations, no validation studies appear available. Results appear more encouraging when sleep diary ratings are compared with PSG variables for a specific sleep episode. Among women followed over multiple nights in the laboratory, overall sleep quality and items reflecting the absence of standard insomnia complaints (difficulty initiating or maintaining sleep, early morning awakening, etc.) were positively associated with PSG-measured sleep efficiency (Akerstedt, Hume, Minors, & Waterhouse, 1994). Ease of awakening, reflecting refreshment or restoration from sleep, was associated with lower sleep efficiency. No relation with sleep stages was demonstrated. Another longitudinal study using sleep diaries found that overall sleep quality was negatively correlated with stage 1 sleep, positively correlated with stage 2 sleep, but unrelated to sleep efficiency among older adults without sleep complaints (O’Donnell et al., 2009). With the potential of aiding validity assessments of epidemiological studies in the general (not clinical) population, the present study aimed to further the understanding of the association
SLEEP QUALITY AND POLYSOMNOGRAPHY CORRELATES
3
Downloaded by [New York University] at 12:37 16 September 2015
between subjective habitual sleep and sleep measured by PSG. We compared sleep quality and restoration from sleep reported in the Karolinska Sleep Questionnaire (KSQ) with subsequent in-home polysomnographic sleep recorded on multiple occasions among individuals without clinical sleep complaints. We hypothesized that subjective sleep quality would be positively related to sleep continuity (e.g., increased sleep efficiency or decreased wake time after sleep onset and number of awakenings) and increased amounts of slow-wave sleep. Because physiological sleep correlates of subjective restoration have been little studied, the analysis was explorative in relation to reports thereof.
METHODS Participants and Design This study is part of a larger project with the overarching aim of examining objective sleep in relation to subjective sleep and health. To this end, volunteers were recruited via advertisements and personal contacts. In total, 52 individuals were approached, and of the 46 that agreed to participate, 13 were lost due to illness, travel, work schedules, and so forth, interfering with the study. Consequently, 33 subjects were included in the study. Data were collected between 1998 and 2000, and all participants (28–69 years, 61% women) were screened by a physician to ensure that they were healthy and did not complain of sleep disturbances. Participants completed the KSQ on a single occasion and underwent multinight PSG recordings in their homes approximately two months later. They received economic compensation equivalent to US$180. The Institutional Review Board at the Karolinska Institutet approved the study. All participants gave written informed consent. Subjective Sleep Measurements The KSQ is a tool to measure subjective habitual sleep and sleepiness, and has mainly been used in Scandinavian studies (see, e.g., Clark et al., 2014; Garde, Albertsen, Persson, Hansen, & Rugulies, 2012; Oyane, Ursin, Pallesen, Holsten, & Bjorvatn, 2008). We investigated two established questionnaire indices reflecting sleep quality and restoration from sleep, respectively. Sleep quality was based on four items: difficulty falling asleep, waking up with difficulty going back to sleep, waking up too early (in the morning), and having restless/disturbed sleep. Restoration from sleep was based on three items: difficulty waking up, waking up feeling unrested, and waking up fatigued. All items referred to the past six months and were rated using a frequency-based format: never, rarely (a few times per year), sometimes (a few times per month), mostly (a few times per week), or always (nearly every day). We assigned responses a value between 5 (never) and 1 (always) and calculated participants’ index scores by averaging responses across the included items. Thus, a higher score indicated less frequent sleep complaints (i.e., better sleep quality or more restoration from sleep). The reliability (internal consistency) of the indices measured with Cronbach’s ˛ has been reported for other nonclinical Swedish samples and was 0.76–0.85 for the sleep quality index (Akerstedt, Ingre, Broman, & Kecklund, 2008; Akerstedt et al., 2002; Hanson et al., 2011; Nordin, Akerstedt, & Nordin, 2013), and 0.73–0.80 for the restoration index (Hanson et al.,
Downloaded by [New York University] at 12:37 16 September 2015
4
WESTERLUND ET AL.
2011; Nordin et al., 2013). The test–retest coefficients were 0.72 and 0.73 for sleep quality and restoration, respectively, for readministration of the included items over a two-year interval (Hanson et al., 2011). Construct validity has been tested on two occasions. First, comparing a representative sample of Swedes to individuals diagnosed with insomnia, it was shown that only 2%–4% of the representative sample had an average sleep quality score as poor as those with insomnia (Akerstedt et al., 2008). Second, correlations between sleep quality and restoration from sleep, respectively, and measures of anxiety, depression, stress, and mental/physical exhaustion were shown to be moderate (0.40–0.60; Nordin et al., 2013), indicating an expected level of homogeneity between these related constructs. Participants also reported usual bedtime (lights out), rise time, and time to falling asleep after lights out (subjective sleep latency) separately for weekdays and weekends. The questionnaire did not assess time spent awake after sleep onset. Hence, subjective total sleep time was defined as the difference between rise time and bedtime, subtracting sleep latency. A weighted average was calculated assigning weekday sleep a weight of 5/7 and weekend sleep a weight of 2/7. Subjective sleep efficiency was calculated as the percentage total sleep time of time in bed after lights out to rise time.
Objective Sleep Recordings Sleep was recorded using unattended home PSG on multiple nonconsecutive nights with a median of 26 days (interquartile range, 18–31 days; range, 7–83 days) between the first and last recording. Most participants (n D 23) completed four sleep recordings, 6 participants completed three recordings, 2 completed five recordings, and 2 completed two recordings. Approximately 50% (n D 16) had at least one recording of weekend sleep. The rationale behind this protocol was to obtain an estimate as representative as possible of habitual physiological sleep. An investigator visited the home of the participants in the evening of each recording day at a prearranged time and attached the electrodes and recording equipment. The participant went to bed at her/his preferred time, pressing the event button of the recorder. In the following morning, a different investigator visited the participants’ home at a prearranged time, removed the electrodes, and collected the equipment. The participants had been instructed how to remove the equipment themselves, should they wish to do so. They had also been advised to refrain from alcohol and vigorous physical activity on the recording days. Recordings were made on Embla devices (Embla Systems, at the time known as Flaga hf, Reykjavik, Iceland) with the following setup: two central electroencephalograms (C3–A2 and C4–A1), a left and right electrooculogram in an oblique pattern at the outer canthi of the eyes, and a chin electromyogram. Ag/AgCl electrodes were used. The sampling rate was set at 100 Hz. To reduce the impact of low-frequency artifacts, a 0.8 Hz high-pass filter was applied for one channel during scoring. This was done for all recordings to ensure that the amount of slow-wave sleep would not be affected. The epoch-to-epoch correlation coefficient with the filter set at 0.5 and 0.8 Hz, respectively, was 0.94 (range, 0.91–0.97), indicating that the total amount of slow-wave sleep (SWS) was not affected. The changes included less stage 4 sleep replaced by slightly more stage 3 sleep and slightly more rapid eye movement (REM) sleep instead of wake. Sleep stages were scored visually in 30-s epochs according to Rechtschaffen and Kales (1968).
SLEEP QUALITY AND POLYSOMNOGRAPHY CORRELATES
5
Standard PSG parameters used in this analysis include sleep efficiency, total sleep time, sleep latency (to stage 1 sleep), stage 1 sleep, stage 2 sleep, slow-wave sleep (stages 3 and 4 combined), REM sleep, wake time after sleep onset, and number of awakenings. Sleep efficiency was calculated as percentage total sleep time of time in bed after lights out (i.e., subtracting sleep latency and wake time after sleep onset from total sleep time) to the time of final awakening. Total sleep time, sleep latency, all sleep stages, and wake time were considered in minutes. Sleep stages were also summarized as percentage total sleep time. For each participant, PSG measures were averaged across the recordings.
Downloaded by [New York University] at 12:37 16 September 2015
Sociodemographic and Lifestyle Data Participants reported weight, height, and smoking status (never/past/current) in a questionnaire. Body mass index (weight in kg/height in meters squared) was calculated. These data are presented for descriptive purposes. Statistical Analysis Numerical variables are described by their median and interquartile range and categorical variables by their frequency and percentage. Differences between subjective and objective sleep-wake times were tested using the Wilcoxon matched-pairs signed-ranks test. The strength of relation between subjective and objective sleep parameters was examined using Spearman’s rank correlation coefficients; these results are reported in Table 1. Significance level was set at 0.05. We used linear regression to model sleep quality and restoration from sleep as a function of the PSG parameters. The questionnaire indices thus represented the outcome variables, and were modeled separately. We found little formal evidence that the outcome variables were not normally distributed. We fit simple and multiple regression models. The former used each PSG parameter separately as predictor variable and the latter used all PSG parameters simultaneously as predictor variables. Total sleep time was relatively strongly correlated with most other PSG parameters, as was sleep efficiency with wake time after sleep onset (Table 1). Hence, to be able to simultaneously consider as many PSG variables as possible and avoid multicollinearity issues, total sleep time and sleep efficiency were excluded from the multipredictor model. The pairwise Spearman’s rank correlation coefficients among the remaining PSG parameters were in the range considered weak to moderate (Table 1). The PSG parameters were primarily modeled as the average of all sleep recordings performed in each participant, and considered in 5-min (% for sleep efficiency) units, except for number of awakenings, which was considered in units of 1. In addition to the average of recordings, we examined whether the variability of PSG data, measured as the standard deviation across recordings, predicted subjective sleep. We also explored the potential effect of age on the results. Age was first included as a continuous predictor variable in the regression analysis alongside any pertinent PSG parameters, that is, PSG parameters consistently associated with the KSQ indices across simple and multiple regression models. If shown to have a substantial impact on the observed results, we dichotomized age at its median (44 years) to examine if the association between KSQ ratings and PSG parameters differed among younger and older individuals.
6
0.38 0.02 0.28 0.61 0.39 0.72 0.10 0.40 0.33 0.00 0.04 0.24 0.34 0.37
0.29 0.29 0.13 0.32 0.46 0.92 0.14 0.15 0.04 0.03 0.04 0.25 0.61 0.21 0.03 0.04 0.07 0.33 0.06 0.29 0.14 0.13 0.14 0.27 0.08 0.12 0.24 0.14 0.46 0.45 0.19 0.17 0.23 0.05 0.14 0.39 0.16 0.16 0.10 0.25 0.17 0.04 0.06 0.63 0.01 0.16
Stage 1 Stage 2 Sleep Sleep
0.37 0.24 0.11 0.16 0.16 0.15 0.27 0.54 0.74
SWS
0.26 0.41 0.18 0.01 0.00 0.06 0.29 0.61
REM Sleep
0.29 0.28 0.01 0.05 0.13 0.16 0.59
Wake Time
0.23 0.08 0.07 0.05 0.04 0.12
0.35 0.23 0.23 0.21 0.14
0.98 0.28 0.33 0.00
0.23 0.31 0.04
0.01 0.24
0.49
Awakenings KSQ-TST KSQ-SE KSQ-latency KSQ-SQ KSQ-RS
Note. KSQ-SQ and KSQ-RS scores range from 1 to 5; higher scores indicate better sleep quality or more restoration from sleep. Statistically significant .p < 0:05/ coefficients are bolded. KSQ, Karolinska Sleep Questionnaire; PSG, polysomnography; REM, rapid eye movement; RS, restoration from sleep; SE, sleep efficiency; SQ, sleep quality; SWS, slow-wave sleep; TST, total sleep time.
PSG-SE PSG-latency Stage 1 sleep Stage 2 sleep SWS REM sleep Wake time Awakenings KSQ-TST KSQ-SE KSQ-latency KSQ-SQ KSQ-RS Age
PSG-TST PSG-SE PSG-latency
TABLE 1 Pairwise Correlations Among Polysomnography Parameters, Habitual Sleep Reported in the Karolinska Sleep Questionnaire, and Age
Downloaded by [New York University] at 12:37 16 September 2015
SLEEP QUALITY AND POLYSOMNOGRAPHY CORRELATES
7
We identified one observation that exerted undue influence on the estimated regression coefficients; it was excluded from analysis. Another participant had missing values on subjective sleep data and also was excluded. Consequently, 31 participants were left for analysis. All analyses were carried out in Stata, version 12 (StataCorp LP, College Station, TX).
Downloaded by [New York University] at 12:37 16 September 2015
RESULTS Table 2 reports the characteristics of the study sample. Sleep quality and restoration from sleep scores were 3.4 (range, 1.5–4.5) and 3.3 (range, 1.7–4.7), respectively; higher scores indicated better sleep quality or more restoration from sleep. Subjective total sleep time was overestimated compared with PSG total sleep time, 443.0 minutes versus 371.0 minutes .p < 0:001/. Subjective sleep efficiency also was higher than its PSG equivalent, 97.8% versus 86.8% .p < 0:001/. Sleep latencies did not differ significantly (subjective, 10.0 min; PSG, 11.9 min; p D 0:52). Percentages of total sleep time spent in each sleep stage were 4.9% for stage 1 sleep, 60.7% for stage 2 sleep, 7.8% for slow-wave sleep, and 26.0% for REM sleep. Median time awake after sleep onset was 48.8 min, and median number of awakenings was 11.8.
TABLE 2 Participant Characteristics Characteristic Age, y Female sex, n (%) Body mass index, kg/m2 Past or current smoking, n (%) KSQ variables Total sleep time, min Sleep efficiency, % Sleep latency, min Sleep quality score Restoration from sleep score Polysomnography variables Total sleep time, min Sleep efficiency, % Sleep latency, min Stage 1 sleep, min Stage 2 sleep, min Slow-wave sleep, min REM sleep, min Wake time after sleep onset, min Awakenings, n
44.0 (33.0–54.0) 20 (60.6) 23.7 (22.6–26.2) 8 (24.0) 443.0 97.8 10.0 3.4 3.3
(427.9–485.0) (94.1–98.7) (6.0–30.0) (2.8–4.0) (3.0–3.7)
371.0 86.8 11.9 17.3 222.1 26.3 100.3 48.8 11.8
(353.8–409.0) (79.8–89.3) (8.0–15.2) (10.5–24.5) (215.3–240.5) (8.8–58.4) (89.5–108.6) (37.3–80.8) (10.8–15.5)
Note. Values represent median (interquartile range) unless otherwise indicated. Scores on sleep quality and restoration from sleep range between 1 and 5; higher scores indicate better sleep quality or more restoration from sleep. KSQ, Karolinska Sleep Questionnaire; REM, rapid eye movement.
8
WESTERLUND ET AL.
Downloaded by [New York University] at 12:37 16 September 2015
Ratings of Habitual Sleep Quality as a Function of PSG Parameters In simple linear regressions, stage 2 sleep was the only significant predictor of subjective sleep quality (ˇ, 0.07; p < 0:01; Table 3; Figure 1). The second strongest predictor was slow-wave sleep (ˇ, 0.05; p D 0:09). Neither sleep efficiency (ˇ, 0.03; p D 0:81) nor total sleep time (ˇ, 0.01; p D 0:29) predicted subjective sleep quality. In the regression model adjusted for multiple PSG parameters, stage 2 sleep remained the most important predictor and the relation with slow-wave sleep reached statistical significance. Although unrelated to sleep quality as single predictors, stage 1 sleep and wake time after sleep onset became significant positive and negative predictors, respectively, in the multiple regression model. Combined, the PSG parameters explained 42% of the variability in sleep quality ratings. We next explored whether the variability of PSG data predicted sleep quality ratings. Overall, the standard deviation of each PSG parameter did not (Table 4). Also, there were no statistically significant correlations between standard deviations and sleep quality (jr j D 0.01–0.35; pvalues > 0.05). Exploring the effect of age on the main results, we included stage 2 sleep, slow-wave sleep, and age simultaneously as predictors of sleep quality. Age did not substantially change the observed associations with the two PSG parameters (data not shown). This was expected since sleep quality was only weakly correlated with age (Table 1).
TABLE 3 Simple and Multiple Linear Regressions of Polysomnography Parameters on Sleep Quality Reported in the Karolinska Sleep Questionnaire Simple Regression
Predictor
ˇ (95% CI)
Sleep efficiency, % Total sleep time, min
0.03 ( 0.22, 0.28) 0.01 ( 0.04, 0.01)
Sleep latency, min Stage 1 sleep, min Stage 2 sleep, min Slow-wave sleep, min REM sleep, min Wake time after sleep onset, min Awakenings, n
0.14 0.03 0.07 0.05 0.01 0.02
( ( ( ( ( (
0.15, 0.12, 0.11, 0.01, 0.08, 0.07,
0.44) 0.19) 0.03) 0.10) 0.06) 0.04)
0.00 ( 0.08, 0.08)
p for Term
Multiple Regression
R2
ˇ (95% CI)
0.81 0.29
0.00 0.04
0.32 0.67
ISSN: 1540-2002 (Print) 1540-2010 (Online) Journal homepage: http://www.tandfonline.com/loi/hbsm20
Relationships Between Questionnaire Ratings of Sleep Quality and Polysomnography in Healthy Adults Anna Westerlund, Ylva Trolle Lagerros, Göran Kecklund, John Axelsson & Torbjörn Åkerstedt To cite this article: Anna Westerlund, Ylva Trolle Lagerros, Göran Kecklund, John Axelsson & Torbjörn Åkerstedt (2014): Relationships Between Questionnaire Ratings of Sleep Quality and Polysomnography in Healthy Adults, Behavioral Sleep Medicine, DOI: 10.1080/15402002.2014.974181 To link to this article: http://dx.doi.org/10.1080/15402002.2014.974181
Published online: 10 Nov 2014.
Submit your article to this journal
Article views: 79
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=hbsm20 Download by: [New York University]
Date: 16 September 2015, At: 12:37
Downloaded by [New York University] at 12:37 16 September 2015
Behavioral Sleep Medicine, 00:1–15, 2014 Copyright © Taylor & Francis Group, LLC ISSN: 1540-2002 print/1540-2010 online DOI: 10.1080/15402002.2014.974181
Relationships Between Questionnaire Ratings of Sleep Quality and Polysomnography in Healthy Adults Anna Westerlund Clinical Epidemiology Unit, Department of Medicine Karolinska Institutet
Ylva Trolle Lagerros Clinical Epidemiology Unit, Department of Medicine Karolinska Institutet
Göran Kecklund Stress Research Institute, Stockholm University Behavioural Science Institute, Radboud University Division of Psychology, Department of Clinical Neuroscience Karolinska Institutet
John Axelsson Division of Psychology, Department of Clinical Neuroscience Karolinska Institutet
Torbjörn Åkerstedt Stress Research Institute, Stockholm University Division of Psychology, Department of Clinical Neuroscience Karolinska Institutet
This study aimed to examine the association between polysomnographic sleep and subjective habitual sleep quality and restoration from sleep. Thirty-one normal sleepers completed the Karolinska Sleep Questionnaire and multiple home polysomnography recordings (n D 2–5). Using linear regression, sleep quality and restoration were separately analyzed as functions of standard Correspondence should be addressed to Anna Westerlund, Karolinska Institutet, Department of Medicine/Solna, Clinical Epidemiology Unit, Eugeniahemmet T2, 171 76, Stockholm, Sweden. E-mail: [email protected]
1
2
WESTERLUND ET AL.
Downloaded by [New York University] at 12:37 16 September 2015
polysomnography parameters: sleep efficiency, total sleep time, sleep latency, stage 1 and 2 sleep, slow-wave sleep, rapid eye movement sleep, wake time after sleep onset, and awakenings (n), averaged across recordings. Stage 2 and slow-wave sleep predicted worse and better sleep quality, respectively. Also, slow-wave sleep predicted less subjective restoration, although adjustment for age attenuated this relation. Our findings lend some physiological validity to ratings of habitual sleep quality in normal sleepers. Data were less supportive of a physiological correlate of ratings of restoration from sleep.
With an ever-growing literature on the link between sleep habits and health outcomes (Ayas et al., 2003; Chandola, Ferrie, Perski, Akbaraly, & Marmot, 2010; Clark, Lange, Hallqvist, Jennum, & Rod, 2014; Gangwisch et al., 2006; Gangwisch, Malaspina, Boden-Albala, & Heymsfield, 2005; Ikehara et al., 2009; Kripke, Garfinkel, Wingard, Klauber, & Marler, 2002; Yaggi, Araujo, & McKinlay, 2006), simple tools to assess habitual sleep are being frequently utilized. Yet, studies on the association of subjective reports of sleep quality and sleep complaints with polysomnographic sleep are scarce. Consequently, validity assessments of epidemiological investigations are also impaired. Habitual sleep is typically measured with single time-point self-administered questionnaires assessing the experience of sleep over some specified period of time. The assumption is that such a report of sleep quality or complaints reflects a trait-like or a long-term state. Examples of self-administered sleep questionnaires include the Pittsburgh Sleep Quality Index (PSQI; Buysse, Reynolds, Monk, Berman, & Kupfer, 1989), the Basic Nordic Sleep Questionnaire (Partinen & Gislason, 1995), the Sleep Quality Scale (Yi, Shin, & Shin, 2006), and the Karolinska Sleep Questionnaire (Kecklund & Åkerstedt, 1992), which were developed to retrospectively assess sleep quality and complaints in various adult populations. Because these questionnaires reflect the subjective experience of sleep, they would likely benefit from validation against polysomnographic (objective) sleep. This has, though, rarely been done. One exception is the validation study of the PSQI among subjects without sleep complaints and depressed or sleep-disorder patients (Buysse et al., 1989). A weak positive correlation between global PSQI scores (a higher score reflecting worse sleep quality) and sleep latency according to polysomnography (PSG) was demonstrated for all subjects. In those with no sleep complaints, the global score positively correlated with percentage sleep time spent in rapid eye movement (REM) sleep. With regard to sleep scales for the assessment of habitual sleep in nonclinical populations, no validation studies appear available. Results appear more encouraging when sleep diary ratings are compared with PSG variables for a specific sleep episode. Among women followed over multiple nights in the laboratory, overall sleep quality and items reflecting the absence of standard insomnia complaints (difficulty initiating or maintaining sleep, early morning awakening, etc.) were positively associated with PSG-measured sleep efficiency (Akerstedt, Hume, Minors, & Waterhouse, 1994). Ease of awakening, reflecting refreshment or restoration from sleep, was associated with lower sleep efficiency. No relation with sleep stages was demonstrated. Another longitudinal study using sleep diaries found that overall sleep quality was negatively correlated with stage 1 sleep, positively correlated with stage 2 sleep, but unrelated to sleep efficiency among older adults without sleep complaints (O’Donnell et al., 2009). With the potential of aiding validity assessments of epidemiological studies in the general (not clinical) population, the present study aimed to further the understanding of the association
SLEEP QUALITY AND POLYSOMNOGRAPHY CORRELATES
3
Downloaded by [New York University] at 12:37 16 September 2015
between subjective habitual sleep and sleep measured by PSG. We compared sleep quality and restoration from sleep reported in the Karolinska Sleep Questionnaire (KSQ) with subsequent in-home polysomnographic sleep recorded on multiple occasions among individuals without clinical sleep complaints. We hypothesized that subjective sleep quality would be positively related to sleep continuity (e.g., increased sleep efficiency or decreased wake time after sleep onset and number of awakenings) and increased amounts of slow-wave sleep. Because physiological sleep correlates of subjective restoration have been little studied, the analysis was explorative in relation to reports thereof.
METHODS Participants and Design This study is part of a larger project with the overarching aim of examining objective sleep in relation to subjective sleep and health. To this end, volunteers were recruited via advertisements and personal contacts. In total, 52 individuals were approached, and of the 46 that agreed to participate, 13 were lost due to illness, travel, work schedules, and so forth, interfering with the study. Consequently, 33 subjects were included in the study. Data were collected between 1998 and 2000, and all participants (28–69 years, 61% women) were screened by a physician to ensure that they were healthy and did not complain of sleep disturbances. Participants completed the KSQ on a single occasion and underwent multinight PSG recordings in their homes approximately two months later. They received economic compensation equivalent to US$180. The Institutional Review Board at the Karolinska Institutet approved the study. All participants gave written informed consent. Subjective Sleep Measurements The KSQ is a tool to measure subjective habitual sleep and sleepiness, and has mainly been used in Scandinavian studies (see, e.g., Clark et al., 2014; Garde, Albertsen, Persson, Hansen, & Rugulies, 2012; Oyane, Ursin, Pallesen, Holsten, & Bjorvatn, 2008). We investigated two established questionnaire indices reflecting sleep quality and restoration from sleep, respectively. Sleep quality was based on four items: difficulty falling asleep, waking up with difficulty going back to sleep, waking up too early (in the morning), and having restless/disturbed sleep. Restoration from sleep was based on three items: difficulty waking up, waking up feeling unrested, and waking up fatigued. All items referred to the past six months and were rated using a frequency-based format: never, rarely (a few times per year), sometimes (a few times per month), mostly (a few times per week), or always (nearly every day). We assigned responses a value between 5 (never) and 1 (always) and calculated participants’ index scores by averaging responses across the included items. Thus, a higher score indicated less frequent sleep complaints (i.e., better sleep quality or more restoration from sleep). The reliability (internal consistency) of the indices measured with Cronbach’s ˛ has been reported for other nonclinical Swedish samples and was 0.76–0.85 for the sleep quality index (Akerstedt, Ingre, Broman, & Kecklund, 2008; Akerstedt et al., 2002; Hanson et al., 2011; Nordin, Akerstedt, & Nordin, 2013), and 0.73–0.80 for the restoration index (Hanson et al.,
Downloaded by [New York University] at 12:37 16 September 2015
4
WESTERLUND ET AL.
2011; Nordin et al., 2013). The test–retest coefficients were 0.72 and 0.73 for sleep quality and restoration, respectively, for readministration of the included items over a two-year interval (Hanson et al., 2011). Construct validity has been tested on two occasions. First, comparing a representative sample of Swedes to individuals diagnosed with insomnia, it was shown that only 2%–4% of the representative sample had an average sleep quality score as poor as those with insomnia (Akerstedt et al., 2008). Second, correlations between sleep quality and restoration from sleep, respectively, and measures of anxiety, depression, stress, and mental/physical exhaustion were shown to be moderate (0.40–0.60; Nordin et al., 2013), indicating an expected level of homogeneity between these related constructs. Participants also reported usual bedtime (lights out), rise time, and time to falling asleep after lights out (subjective sleep latency) separately for weekdays and weekends. The questionnaire did not assess time spent awake after sleep onset. Hence, subjective total sleep time was defined as the difference between rise time and bedtime, subtracting sleep latency. A weighted average was calculated assigning weekday sleep a weight of 5/7 and weekend sleep a weight of 2/7. Subjective sleep efficiency was calculated as the percentage total sleep time of time in bed after lights out to rise time.
Objective Sleep Recordings Sleep was recorded using unattended home PSG on multiple nonconsecutive nights with a median of 26 days (interquartile range, 18–31 days; range, 7–83 days) between the first and last recording. Most participants (n D 23) completed four sleep recordings, 6 participants completed three recordings, 2 completed five recordings, and 2 completed two recordings. Approximately 50% (n D 16) had at least one recording of weekend sleep. The rationale behind this protocol was to obtain an estimate as representative as possible of habitual physiological sleep. An investigator visited the home of the participants in the evening of each recording day at a prearranged time and attached the electrodes and recording equipment. The participant went to bed at her/his preferred time, pressing the event button of the recorder. In the following morning, a different investigator visited the participants’ home at a prearranged time, removed the electrodes, and collected the equipment. The participants had been instructed how to remove the equipment themselves, should they wish to do so. They had also been advised to refrain from alcohol and vigorous physical activity on the recording days. Recordings were made on Embla devices (Embla Systems, at the time known as Flaga hf, Reykjavik, Iceland) with the following setup: two central electroencephalograms (C3–A2 and C4–A1), a left and right electrooculogram in an oblique pattern at the outer canthi of the eyes, and a chin electromyogram. Ag/AgCl electrodes were used. The sampling rate was set at 100 Hz. To reduce the impact of low-frequency artifacts, a 0.8 Hz high-pass filter was applied for one channel during scoring. This was done for all recordings to ensure that the amount of slow-wave sleep would not be affected. The epoch-to-epoch correlation coefficient with the filter set at 0.5 and 0.8 Hz, respectively, was 0.94 (range, 0.91–0.97), indicating that the total amount of slow-wave sleep (SWS) was not affected. The changes included less stage 4 sleep replaced by slightly more stage 3 sleep and slightly more rapid eye movement (REM) sleep instead of wake. Sleep stages were scored visually in 30-s epochs according to Rechtschaffen and Kales (1968).
SLEEP QUALITY AND POLYSOMNOGRAPHY CORRELATES
5
Standard PSG parameters used in this analysis include sleep efficiency, total sleep time, sleep latency (to stage 1 sleep), stage 1 sleep, stage 2 sleep, slow-wave sleep (stages 3 and 4 combined), REM sleep, wake time after sleep onset, and number of awakenings. Sleep efficiency was calculated as percentage total sleep time of time in bed after lights out (i.e., subtracting sleep latency and wake time after sleep onset from total sleep time) to the time of final awakening. Total sleep time, sleep latency, all sleep stages, and wake time were considered in minutes. Sleep stages were also summarized as percentage total sleep time. For each participant, PSG measures were averaged across the recordings.
Downloaded by [New York University] at 12:37 16 September 2015
Sociodemographic and Lifestyle Data Participants reported weight, height, and smoking status (never/past/current) in a questionnaire. Body mass index (weight in kg/height in meters squared) was calculated. These data are presented for descriptive purposes. Statistical Analysis Numerical variables are described by their median and interquartile range and categorical variables by their frequency and percentage. Differences between subjective and objective sleep-wake times were tested using the Wilcoxon matched-pairs signed-ranks test. The strength of relation between subjective and objective sleep parameters was examined using Spearman’s rank correlation coefficients; these results are reported in Table 1. Significance level was set at 0.05. We used linear regression to model sleep quality and restoration from sleep as a function of the PSG parameters. The questionnaire indices thus represented the outcome variables, and were modeled separately. We found little formal evidence that the outcome variables were not normally distributed. We fit simple and multiple regression models. The former used each PSG parameter separately as predictor variable and the latter used all PSG parameters simultaneously as predictor variables. Total sleep time was relatively strongly correlated with most other PSG parameters, as was sleep efficiency with wake time after sleep onset (Table 1). Hence, to be able to simultaneously consider as many PSG variables as possible and avoid multicollinearity issues, total sleep time and sleep efficiency were excluded from the multipredictor model. The pairwise Spearman’s rank correlation coefficients among the remaining PSG parameters were in the range considered weak to moderate (Table 1). The PSG parameters were primarily modeled as the average of all sleep recordings performed in each participant, and considered in 5-min (% for sleep efficiency) units, except for number of awakenings, which was considered in units of 1. In addition to the average of recordings, we examined whether the variability of PSG data, measured as the standard deviation across recordings, predicted subjective sleep. We also explored the potential effect of age on the results. Age was first included as a continuous predictor variable in the regression analysis alongside any pertinent PSG parameters, that is, PSG parameters consistently associated with the KSQ indices across simple and multiple regression models. If shown to have a substantial impact on the observed results, we dichotomized age at its median (44 years) to examine if the association between KSQ ratings and PSG parameters differed among younger and older individuals.
6
0.38 0.02 0.28 0.61 0.39 0.72 0.10 0.40 0.33 0.00 0.04 0.24 0.34 0.37
0.29 0.29 0.13 0.32 0.46 0.92 0.14 0.15 0.04 0.03 0.04 0.25 0.61 0.21 0.03 0.04 0.07 0.33 0.06 0.29 0.14 0.13 0.14 0.27 0.08 0.12 0.24 0.14 0.46 0.45 0.19 0.17 0.23 0.05 0.14 0.39 0.16 0.16 0.10 0.25 0.17 0.04 0.06 0.63 0.01 0.16
Stage 1 Stage 2 Sleep Sleep
0.37 0.24 0.11 0.16 0.16 0.15 0.27 0.54 0.74
SWS
0.26 0.41 0.18 0.01 0.00 0.06 0.29 0.61
REM Sleep
0.29 0.28 0.01 0.05 0.13 0.16 0.59
Wake Time
0.23 0.08 0.07 0.05 0.04 0.12
0.35 0.23 0.23 0.21 0.14
0.98 0.28 0.33 0.00
0.23 0.31 0.04
0.01 0.24
0.49
Awakenings KSQ-TST KSQ-SE KSQ-latency KSQ-SQ KSQ-RS
Note. KSQ-SQ and KSQ-RS scores range from 1 to 5; higher scores indicate better sleep quality or more restoration from sleep. Statistically significant .p < 0:05/ coefficients are bolded. KSQ, Karolinska Sleep Questionnaire; PSG, polysomnography; REM, rapid eye movement; RS, restoration from sleep; SE, sleep efficiency; SQ, sleep quality; SWS, slow-wave sleep; TST, total sleep time.
PSG-SE PSG-latency Stage 1 sleep Stage 2 sleep SWS REM sleep Wake time Awakenings KSQ-TST KSQ-SE KSQ-latency KSQ-SQ KSQ-RS Age
PSG-TST PSG-SE PSG-latency
TABLE 1 Pairwise Correlations Among Polysomnography Parameters, Habitual Sleep Reported in the Karolinska Sleep Questionnaire, and Age
Downloaded by [New York University] at 12:37 16 September 2015
SLEEP QUALITY AND POLYSOMNOGRAPHY CORRELATES
7
We identified one observation that exerted undue influence on the estimated regression coefficients; it was excluded from analysis. Another participant had missing values on subjective sleep data and also was excluded. Consequently, 31 participants were left for analysis. All analyses were carried out in Stata, version 12 (StataCorp LP, College Station, TX).
Downloaded by [New York University] at 12:37 16 September 2015
RESULTS Table 2 reports the characteristics of the study sample. Sleep quality and restoration from sleep scores were 3.4 (range, 1.5–4.5) and 3.3 (range, 1.7–4.7), respectively; higher scores indicated better sleep quality or more restoration from sleep. Subjective total sleep time was overestimated compared with PSG total sleep time, 443.0 minutes versus 371.0 minutes .p < 0:001/. Subjective sleep efficiency also was higher than its PSG equivalent, 97.8% versus 86.8% .p < 0:001/. Sleep latencies did not differ significantly (subjective, 10.0 min; PSG, 11.9 min; p D 0:52). Percentages of total sleep time spent in each sleep stage were 4.9% for stage 1 sleep, 60.7% for stage 2 sleep, 7.8% for slow-wave sleep, and 26.0% for REM sleep. Median time awake after sleep onset was 48.8 min, and median number of awakenings was 11.8.
TABLE 2 Participant Characteristics Characteristic Age, y Female sex, n (%) Body mass index, kg/m2 Past or current smoking, n (%) KSQ variables Total sleep time, min Sleep efficiency, % Sleep latency, min Sleep quality score Restoration from sleep score Polysomnography variables Total sleep time, min Sleep efficiency, % Sleep latency, min Stage 1 sleep, min Stage 2 sleep, min Slow-wave sleep, min REM sleep, min Wake time after sleep onset, min Awakenings, n
44.0 (33.0–54.0) 20 (60.6) 23.7 (22.6–26.2) 8 (24.0) 443.0 97.8 10.0 3.4 3.3
(427.9–485.0) (94.1–98.7) (6.0–30.0) (2.8–4.0) (3.0–3.7)
371.0 86.8 11.9 17.3 222.1 26.3 100.3 48.8 11.8
(353.8–409.0) (79.8–89.3) (8.0–15.2) (10.5–24.5) (215.3–240.5) (8.8–58.4) (89.5–108.6) (37.3–80.8) (10.8–15.5)
Note. Values represent median (interquartile range) unless otherwise indicated. Scores on sleep quality and restoration from sleep range between 1 and 5; higher scores indicate better sleep quality or more restoration from sleep. KSQ, Karolinska Sleep Questionnaire; REM, rapid eye movement.
8
WESTERLUND ET AL.
Downloaded by [New York University] at 12:37 16 September 2015
Ratings of Habitual Sleep Quality as a Function of PSG Parameters In simple linear regressions, stage 2 sleep was the only significant predictor of subjective sleep quality (ˇ, 0.07; p < 0:01; Table 3; Figure 1). The second strongest predictor was slow-wave sleep (ˇ, 0.05; p D 0:09). Neither sleep efficiency (ˇ, 0.03; p D 0:81) nor total sleep time (ˇ, 0.01; p D 0:29) predicted subjective sleep quality. In the regression model adjusted for multiple PSG parameters, stage 2 sleep remained the most important predictor and the relation with slow-wave sleep reached statistical significance. Although unrelated to sleep quality as single predictors, stage 1 sleep and wake time after sleep onset became significant positive and negative predictors, respectively, in the multiple regression model. Combined, the PSG parameters explained 42% of the variability in sleep quality ratings. We next explored whether the variability of PSG data predicted sleep quality ratings. Overall, the standard deviation of each PSG parameter did not (Table 4). Also, there were no statistically significant correlations between standard deviations and sleep quality (jr j D 0.01–0.35; pvalues > 0.05). Exploring the effect of age on the main results, we included stage 2 sleep, slow-wave sleep, and age simultaneously as predictors of sleep quality. Age did not substantially change the observed associations with the two PSG parameters (data not shown). This was expected since sleep quality was only weakly correlated with age (Table 1).
TABLE 3 Simple and Multiple Linear Regressions of Polysomnography Parameters on Sleep Quality Reported in the Karolinska Sleep Questionnaire Simple Regression
Predictor
ˇ (95% CI)
Sleep efficiency, % Total sleep time, min
0.03 ( 0.22, 0.28) 0.01 ( 0.04, 0.01)
Sleep latency, min Stage 1 sleep, min Stage 2 sleep, min Slow-wave sleep, min REM sleep, min Wake time after sleep onset, min Awakenings, n
0.14 0.03 0.07 0.05 0.01 0.02
( ( ( ( ( (
0.15, 0.12, 0.11, 0.01, 0.08, 0.07,
0.44) 0.19) 0.03) 0.10) 0.06) 0.04)
0.00 ( 0.08, 0.08)
p for Term
Multiple Regression
R2
ˇ (95% CI)
0.81 0.29
0.00 0.04
0.32 0.67
